[https://nvbugs/6368053][fix] (1) Broaden the assertion to is_eagle3() or is_eagle3_one_model() - #15689
Closed
tensorrt-cicd wants to merge 1 commit into
Closed
[https://nvbugs/6368053][fix] (1) Broaden the assertion to is_eagle3() or is_eagle3_one_model()#15689tensorrt-cicd wants to merge 1 commit into
is_eagle3() or is_eagle3_one_model()#15689tensorrt-cicd wants to merge 1 commit into
Conversation
…n, sync runtime_draft_len The static-tree branch of TrtllmAttentionMetadata.update_spec_dec_param asserted spec_metadata.spec_dec_mode.is_eagle3(), which only matches EAGLE3 (two-model). With quickstart_advanced.py defaulting use_one_model=True, eagle_choices triggers the one-model EAGLE3 path whose spec_dec_mode is EAGLE3_ONE_MODEL, so the assertion fired during warmup. Broaden the guard to accept both Eagle3 variants — they both set is_spec_dec_tree=True from eagle_choices and the downstream spec_tree_manager-based logic is the same. Also sync self.runtime_draft_len to max_total_draft_tokens in forward() for non-linear-tree configs, mirroring the update that _prepare_tp_inputs performs later. The attention-warmup path (_run_attention_warmup -> forward) bypasses _prepare_tp_inputs's update, so spec_metadata.runtime_draft_len was stale (=max_draft_len) while the draft_tokens buffer was sized for max_total_draft_tokens. This caused "shape '[1, 1]' is invalid for input of size 3" in sample_and_accept_draft_tokens once the original assertion was relaxed. Remove the two test_draft_token_tree_quickstart_advanced_eagle3* entries from waives.txt now that both tests pass. Signed-off-by: tensorrt-cicd <90828364+tensorrt-cicd@users.noreply.github.com>
Contributor
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Enterprise Run ID: 📒 Files selected for processing (3)
💤 Files with no reviewable changes (1)
📝 WalkthroughWalkthroughBroadens the static-tree spec-decoding mode assertion in ChangesEagle3OneModel spec-dec fixes
Estimated code review effort🎯 2 (Simple) | ⏱️ ~10 minutes Possibly related PRs
Suggested reviewers
🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
Collaborator
|
/bot run --disable-fail-fast |
Collaborator
Author
|
PR_Github #58681 [ run ] triggered by Bot. Commit: |
Collaborator
Author
|
PR_Github #58681 [ run ] completed with state
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
is_eagle3() or is_eagle3_one_model(). (2) In model_engine.forward(), pre-set self.runtime_draft_len=max_total_draft_tokens for non-linear-tree spec configs before syncing spec_metadata.runtime_draft_len, mirroring _prepare_tp_inputs. (3) Remove the two now-passing waivers under nvbugs/6368053.Test plan
Links
Summary by CodeRabbit